Examination of Signal-to-Noise Ratio in Remote Photoplethysmography Signal Extraction for Video Authenticity Detection

Remote photoplethysmography (rPPG) is a non-invasive technique for assessing physiological signals such as heart rate through facial video recordings. This research investi-gates the impact of signal-to-noise ratio (SNR) on selecting rPPG signals from the face region and examines how additional metrics like Power Spectral Density (PSD), Standard Deviation (STD), and Mean Absolute Deviation (MAD) contribute to differentiating real and fake videos. The results show that while higher SNR values correlate with more accurate rPPG signal extraction, SNR alone is insufficient for distinguishing authentic and counterfeit videos due to similar SNR distributions. By combining PSD, STD, and MAD, we analyzed key facial regions of interest (ROIs) and implemented in Support Vector Machine (SVM) classifier, achieving an accuracy of 98 %. This demonstrates the effectiveness of these metrics in improving accuracy by filtering noise and capturing consistent physiological patterns for detecting deepfakes.